AGENT DONE
A bounded artifact reaches the gate.
Independent evidence gate
GenXis GavelTM independently checks bounded evidence before work counts as done. Exact decision. Inspectable evidence. Signed receipt.

From install to first proof
01 / one command · one email · no passwordnpm i genxis-gavelThe verification path
A bounded artifact reaches the gate.
Independent checks reproduce the claim.
An exact decision, never an invented pass.
The result is bound to its evidence root.
Developer workflow
# Install npm i genxis-gavel # Confirm readiness npx gavel status --json # Check one artifact npx gavel check .\change.patch --json ` --receipt-out .\gavel-receipt.json
Why we built it
We built GavelTM from the frustration of watching agent work arrive with confident completion claims, unclear evidence, and too much human time spent finding failures after the fact.
We wanted a separate authority inside the loop, not another model grading itself. The goal is to help more honest, better-behaved AI systems approach enterprise-class performance and reliability by making evidence, correction, and no-regression decisions part of how work gets finished.
Verified in the wild
311Certified catches this week
Rolling seven-day total from all distinct Most Annoying AI benchmark receipt roots plus the production catch ledger.This week · refreshes automatically
No card is needed to begin. Every pass stays free. Your magic-link account pays nothing for its first ten independently certified catches.
The first ten certified catches on each new magic-link account.
Clean outcomes stay free, before and after the launch allowance.
A verifier-certified medium, low, informational, or minor finding.
A verifier-certified high, critical, major, or material finding.
A bounded defect with localized impact, certified at medium, low, informational, or minor severity by the verifier responsible for that domain.
A high, critical, major, or material defect with independently reproduced evidence and a receipt bound to the checked artifact.
The current public deterministic security mesh contains high and critical predicates, so its certified findings enter the 50¢ tier. The 10¢ tier applies when an enabled domain verifier emits a certified medium, low, informational, or minor finding. These examples describe bounded predicates, not complete security coverage.
Malformed, duplicate, timed-out, denied, or unverified outcomes are not billed. Only independently reproduced catches count toward the launch allowance.
Claim ceiling
GavelTM proves only the bounded checks named in its receipt. It makes failures and uncertainty explicit instead of stretching a result into a larger promise.
Catch boundaries
A reusable predicate that can apply across many repositories or domains: a recognized secret format, shell-backed execution, disabled TLS verification, or another precisely declared pattern.
General reach does not make it certified. The candidate must still be reproduced by its verifier and bound to the submitted evidence.
A finding judged inside a declared domain contract: a frozen benchmark answer, schema, policy, regression command, test oracle, or other bounded source of truth.
The GavelTM decision proves only that domain predicate for that evidence. It does not automatically generalize outside the contract.
Fail-closed answer: those outcomes may receive an explicit denied, invalid, or unverified status, but they do not become a certified catch, a pass, or a promotion. No evidence never means success.
Your first check
We’ll email a single-use link. No password, no credential stored in this browser.
If that address can continue, a single-use link is on its way. You may close this page.
Exchanging the single-use link securely.
Copy it now. It is shown only in this page and is never persisted by this site.
gavel status --jsonIt may be expired or already used. Request a fresh link to continue.